Author: Georg v. Zimmermann
Date: 04:57:10 10/03/01
Go up one level in this thread
3 more comments: for root processors using hash entries from the last search or from pondering is "incorrect", thats another big problem. Many programs are root processors even if they only decide on which state of the game (middle game, early end game ...) they are in at the root. Some do little root processing, a lot in the first plys of the search and only little expensive eval() at the leaves. Georg On October 02, 2001 at 08:38:01, Steve Maughan wrote: >Gordon, > >Some programs do a 'thorough' static analysis of the position before they start >to search. They then assign values for a piece being on any one of the squares. > This is root node processing or pre-processing. The advantage is speed of >evaluation but the (big) dissadvantage is that what seems like a good idea at >the root may not be so good at the tip of a search e.g. rook on 7th when King is >on 8th is good but when the king starts to advance up the board the preprocessor > still thinks that the rook on the 7th is good. Moreover, as machines get >faster and search deeper the problem gets worse since the tips are further away >from the root. > >The alternative is tip evaluation which is slower but *MUCH* more accurate when >done well. The famous pre-processors are early Fritz versions, Rex chess, >Travel Champion, Novag Constellation and anything that claims to do >1.5 million >nodes / sec on a single processor e.g. Little Goliath. Junior was always >considered to be a preprocessor but I seriously doubt it now as the evaluation >is usually rock solid between moves and the scores seem to contain extensive >king safety elements. > >Steve
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.